src/panfrost/bifrost/bi_ra.c

   1 /*
   2  * Copyright (C) 2020 Collabora Ltd.
   3  *
   4  * Permission is hereby granted, free of charge, to any person obtaining a
   5  * copy of this software and associated documentation files (the "Software"),
   6  * to deal in the Software without restriction, including without limitation
   7  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
   8  * and/or sell copies of the Software, and to permit persons to whom the
   9  * Software is furnished to do so, subject to the following conditions:
  10  *
  11  * The above copyright notice and this permission notice (including the next
  12  * paragraph) shall be included in all copies or substantial portions of the
  13  * Software.
  14  *
  15  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  16  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  17  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
  18  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  19  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
  20  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  21  * SOFTWARE.
  22  *
  23  * Authors (Collabora):
  24  *      Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
  25  */
  26
  27 #include "compiler.h"
  28 #include "panfrost/util/lcra.h"
  29 #include "util/u_memory.h"
  30
  31 static void
  32 bi_compute_interference(bi_context *ctx, struct lcra_state *l)
  33 {
  34         bi_compute_liveness(ctx);
  35
  36         bi_foreach_block(ctx, _blk) {
  37                 bi_block *blk = (bi_block *) _blk;
  38                 uint16_t *live = mem_dup(_blk->live_out, l->node_count * sizeof(uint16_t));
  39
  40                 bi_foreach_instr_in_block_rev(blk, ins) {
  41                         /* Mark all registers live after the instruction as
  42                          * interfering with the destination */
  43
  44                         if (ins->dest && (ins->dest < l->node_count)) {
  45                                 for (unsigned i = 1; i < l->node_count; ++i) {
  46                                         if (live[i])
  47                                                 lcra_add_node_interference(l, ins->dest, ins->writemask, i, live[i]);
  48                                 }
  49                         }
  50
  51                         /* Update live_in */
  52                         bi_liveness_ins_update(live, ins, l->node_count);
  53                 }
  54
  55                 free(live);
  56         }
  57 }
  58
  59 enum {
  60         BI_REG_CLASS_WORK = 0,
  61 } bi_reg_class;
  62
  63 static struct lcra_state *
  64 bi_allocate_registers(bi_context *ctx, bool *success)
  65 {
  66         unsigned node_count = bi_max_temp(ctx);
  67
  68         struct lcra_state *l =
  69                 lcra_alloc_equations(node_count, 1, 8, 16, 1);
  70
  71         l->class_start[BI_REG_CLASS_WORK] = 0;
  72         l->class_size[BI_REG_CLASS_WORK] = 64 * 4; /* R0 - R63, all 32-bit */
  73
  74         bi_foreach_instr_global(ctx, ins) {
  75                 unsigned dest = ins->dest;
  76
  77                 if (!dest || (dest >= node_count))
  78                         continue;
  79
  80                 l->class[dest] = BI_REG_CLASS_WORK;
  81                 lcra_set_alignment(l, dest, 2); /* 2^2 = 4 */
  82                 lcra_restrict_range(l, dest, 4);
  83         }
  84
  85         bi_compute_interference(ctx, l);
  86
  87         *success = lcra_solve(l);
  88
  89         return l;
  90 }
  91
  92 static unsigned
  93 bi_reg_from_index(struct lcra_state *l, unsigned index, unsigned offset)
  94 {
  95         /* Did we run RA for this index at all */
  96         if (index >= l->node_count)
  97                 return index;
  98
  99         /* LCRA didn't bother solving this index (how lazy!) */
 100         signed solution = l->solutions[index];
 101         if (solution < 0)
 102                 return index;
 103
 104         solution += offset;
 105
 106         assert((solution & 0x3) == 0);
 107         unsigned reg = solution / 4;
 108         return BIR_INDEX_REGISTER | reg;
 109 }
 110
 111 static void
 112 bi_adjust_src_ra(bi_instruction *ins, struct lcra_state *l, unsigned src)
 113 {
 114         if (ins->src[src] >= l->node_count)
 115                 return;
 116
 117         bool vector = (bi_class_props[ins->type] & BI_VECTOR) && src == 0;
 118         unsigned offset = 0;
 119
 120         if (vector) {
 121                 /* TODO: Do we do anything here? */
 122         } else {
 123                 /* Use the swizzle as component select */
 124                 nir_alu_type T = ins->src_types[src];
 125                 unsigned size = nir_alu_type_get_type_size(T);
 126                 assert(size <= 32); /* TODO: 64-bit */
 127                 unsigned comps_per_reg = 32 / size;
 128                 unsigned components = bi_get_component_count(ins, src);
 129
 130                 for (unsigned i = 0; i < components; ++i) {
 131                         unsigned off = ins->swizzle[src][i] / comps_per_reg;
 132                         off *= 4; /* 32-bit registers */
 133
 134                         /* We can't cross register boundaries in a swizzle */
 135                         if (i == 0)
 136                                 offset = off;
 137                         else
 138                                 assert(off == offset);
 139
 140                         ins->swizzle[src][i] %= comps_per_reg;
 141                 }
 142         }
 143
 144         ins->src[src] = bi_reg_from_index(l, ins->src[src], offset);
 145 }
 146
 147 static void
 148 bi_adjust_dest_ra(bi_instruction *ins, struct lcra_state *l)
 149 {
 150         if (ins->dest >= l->node_count)
 151                 return;
 152
 153         bool vector = (bi_class_props[ins->type] & BI_VECTOR);
 154         unsigned offset = 0;
 155
 156         if (!vector) {
 157                 /* Look at the writemask to get an offset, specifically the
 158                  * trailing zeros */
 159
 160                 unsigned tz = __builtin_ctz(ins->writemask);
 161
 162                 /* Recall writemask is one bit per byte, so tz is in eytes */
 163                 unsigned regs = tz / 4;
 164                 offset = regs * 4;
 165
 166                 /* Adjust writemask to compensate */
 167                 ins->writemask >>= offset;
 168         }
 169
 170         ins->dest = bi_reg_from_index(l, ins->dest, offset);
 171
 172 }
 173
 174 static void
 175 bi_install_registers(bi_context *ctx, struct lcra_state *l)
 176 {
 177         bi_foreach_instr_global(ctx, ins) {
 178                 bi_adjust_dest_ra(ins, l);
 179
 180                 bi_foreach_src(ins, s)
 181                         bi_adjust_src_ra(ins, l, s);
 182         }
 183 }
 184
 185 void
 186 bi_register_allocate(bi_context *ctx)
 187 {
 188         struct lcra_state *l = NULL;
 189         bool success = false;
 190
 191         do {
 192                 if (l) {
 193                         lcra_free(l);
 194                         l = NULL;
 195                 }
 196
 197                 bi_invalidate_liveness(ctx);
 198                 l = bi_allocate_registers(ctx, &success);
 199
 200                 /* TODO: Spilling */
 201                 assert(success);
 202         } while(!success);
 203
 204         bi_install_registers(ctx, l);
 205
 206         lcra_free(l);
 207 }